Create, manage, and execute agent tools (claude, codex) inside Docker sandboxes for isolated code execution. Use when running agent loops, spawning tool subprocesses, or any task requiring process isolation. Triggers on "sandbox", "isolated execution", "docker sandbox", "safe agent execution", or when working on agent loop infrastructure.
92
88%
Does it follow best practices?
Impact
100%
3.22xAverage score across 3 eval scenarios
Risky
Do not use without reviewing
Agent loop sandbox lifecycle management
Sandbox naming convention
41%
100%
createLoopSandbox function
25%
100%
execInSandbox function
62%
100%
destroyLoopSandbox function
87%
100%
Pre-warm create at loop start
100%
100%
Reuse across stories
100%
100%
Destroy at loop end
100%
100%
spawnTool sandbox check
20%
100%
Host-mode fallback
100%
100%
Working dir flag
75%
100%
Exec overhead documented
40%
100%
Auth token storage and sandbox injection
Claude token command
0%
100%
Claude secret name
0%
100%
Secrets lease TTL flag
0%
100%
CLAUDE_CODE_OAUTH_TOKEN env var
0%
100%
Codex secret name
0%
100%
Codex auth source
0%
100%
Codex sandbox injection path
0%
100%
Codex portability noted
0%
100%
Claude token validity documented
0%
100%
Refresh procedures documented
0%
100%
Network restriction and custom sandbox templates
Bun absent from base template
41%
100%
Network deny policy
0%
100%
Allow-host for API endpoints
0%
100%
Both required hosts allowed
40%
100%
Save template command
0%
100%
Template versioning
87%
100%
Launch from template (-t flag)
0%
100%
Template reuse rationale
87%
100%
Bun install method
62%
100%
825972c
Table of Contents
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.